Tensor Decomposition of Stimulated Monocyte and Macrophage Gene Expression Profiles Identifies Neurodegenerative Disease-specific Trans-eQTLs

Abstract

Recent human genetic studies suggest that cells of the innate immune system have a primary role in the pathogenesis of neurodegenerative diseases. However, the results from these studies often do not elucidate how the genetic variants affect the biology of these cells to modulate disease risk. Here, we applied a tensor decomposition method to uncover disease-associated gene networks linked to distal genetic variation in stimulated human monocytes and macrophages gene expression profiles. We report robust evidence that some disease-associated genetic variants affect the expression of multiple genes in trans. These include a Parkinson's disease locus influencing the expression of genes mediated by a protease that controls lysosomal function, and Alzheimer's disease loci influencing the expression of genes involved in type 1 interferon signaling, myeloid phagocytosis, and complement cascade pathways. Overall, we uncover gene networks in induced innate immune cells linked to disease-associated genetic variants, which may help elucidate the underlying biology of disease.

Co-expression analysis reveals interpretable gene modules controlled by trans-acting genetic variants

“…Extensive follow-up with gene set and transcription factor motif enrichment analyses allowed us to gain additional insight into the functional impact of trans -eQTLs and prioritise loci for further analyses. In addition to replicating two known monocyte-specific trans -eQTLs at the IFNB1 ( Fairfax et al, 2014 ; Quach et al, 2016 ; Ramdhani et al, 2020 ; Ruffieux et al, 2018 ) and LYZ loci ( Fairfax et al, 2012 ; Rakitsch and Stegle, 2016 ; Rotival et al, 2011 ), we found that the trans -eQTL at the ARHGEF3 locus detected in multiple whole blood datasets ( Mao et al, 2019 ; Nath et al, 2017 ; Rotival et al, 2011 ; Wheeler et al, 2019 ) was highly specific to platelets in our analysis. Finally, we also detected a novel association at the SLC39A8 locus that controlled a group of genes encoding zinc-binding proteins in LPS-stimulated monocytes.…”

“…To reduce the number of tested phenotypes, co-expression analysis methods are sometimes used to aggregate individual genes to co-expressed modules capturing signalling pathways and cellular processes ( Stein-O'Brien et al, 2018 ). Such approaches have been successful in identifying trans -eQTLs in yeast ( Parts et al, 2011 ) as well as various human tissues ( Hore et al, 2016 ; Mao et al, 2019 ; Nath et al, 2017 ) and purified immune cells ( Ramdhani et al, 2020 ; Rotival et al, 2011 ). An added benefit of co-expression modules is that they can often be directly interpreted as signatures of higher level cellular phenotypes, such as activation of specific signalling pathways or transcription factors ( Parts et al, 2011 ; Way et al, 2020 ).…”

“…Recently, both matrix factorisation and co-expression clustering methods have been further extended to incorporate prior information about biological pathways and gene sets, resulting in pathway-level information extractor (PLIER) ( Mao et al, 2019 ) and funcExplorer ( Kolberg et al, 2018 ), respectively. Out of these methods, ICA, WGCNA, SDA and PLIER have previously been used to find trans -eQTLs for modules of co-expressed genes ( Hore et al, 2016 ; Mao et al, 2019 ; Nath et al, 2017 ; Ramdhani et al, 2020 ; Rotival et al, 2011 ), but only a single method at a time. However, since different methods solve distinct optimisation problems, they can detect complementary sets of co-expression modules ( Stein-O'Brien et al, 2018 ), with recent benchmarks demonstrating that there is no single best co-expression analysis method ( Way et al, 2020 ).…”

A unified model of dementias and age‐related neurodegeneration

“…This is in line with the current emphasis on regulatory rather than coding variables, that is, EQTLs (expression quantitative trait loci), particularly in the context of age-related dementias. 100,101 A natural implication of the central role of epigenetic shifts is that both the cell's underlying genetics and its prior epigenetic state will strongly affect the functional outcome of these epigenetic shifts. Because of different genes, different species will express different cellular outcomes despite equivalent changes in telomere lengths.…”

Identification of genes associated with altered gene expression and m6A profiles during hypoxia using tensor decomposition based unsupervised feature extraction

“…There are a limited number of genomic studies using TD 9 , 10 . Fang proposed tightly integrated genomic and epigenomic data mining using TD 11 (445 samples for TCGA-OV and 480 samples for TCGA-HNSC), Hore et al applied TD to multi-tissue gene expression experiments 12 (845 related individuals), Ramdhani et al applied TD to stimulated monocyte and macrophage gene expression profiles 13 (432 samples), Wang et al applied TD to multi-tissue multi-individual gene expression 14 (544 individuals), Li et al applied TD to clinical gene-sample-time microarray expression 15 (53 genes and 27 samples), Hu et al applied TD to gene expression of tumor samples 16 (more than 11,000 tumor samples), Diaz et al applied TD to genomic data 17 (503 patients), and Bradley et al applied TD to DNA copy-number alterations 18 (a few hundred samples). All methods other than that used by Li et al included as few as 53 genes and required as many as several hundred samples, whereas our methods generally require only a few samples (in this study as few as eight samples).…”